Rémi Denis-Courmont wrote:
Hello,
I take the freedom to move this to 6man mailing list...
Thanks, please see below.
On Thu, 02 Jul 2009 19:30:54 +0100, Gorry Fairhurst <[email protected]>
wrote:
Have you any comments on:
draft-fairhurst-6man-tsvwg-udptt-01
I am trying to figure-out whether this is a good thing to work further
on.
Canonically, you would use UDP-Lite for this. Of course, UDP-Lite is
impractical in IPv4 due to incompatible middleboxes, especially NAPTs. But
as it comes to IPv6, I do not see this as a real problem with UDP-Lite over
*IPv6* (contrary to IPv4).
I also suggested this and if UDP-Lite satisfies the need, I'd be
happy to see this used - I would see that increasing UDP-Lite support
has wider benefit.
Admittedly, UDP-Lite over IPv6 will still have problem, should there be
v4-v6 protocol translation (PT) on the path (I think that was Dave Thaler's
counter-argument at the last meeting). I am not convinced that this is a
serious case here, as it seems dubious to me that we will see UDPTT through
protocol translators. But lets say we will.
Then I would like to know why and how UDPTT will work better than UDP-Lite.
Namely, I would like to know how UDPTT/IPv6<->checksum-free UDP/IPv4
translation is supposed to happen, why IPv4 middleboxes won't explode on
the resulting packets, and how it solves the checksum performance issue.
UDPTT results in a static checksum per flow, which helps reduce work at
the sender.
It also overloads the IPv6 next header for UDP (136 decimal), which
gives it some chance of operation in already deployed systems - but one
of the key features is that it does not have the unwanted side-effects
of zero'ing the IPv6 UDP checksum. UDPTT is a smaller step, than UDP-Lite.
I am not in any way implying that this is possible or impossible or anything
- but I think it belongs in the draft or in a companion draft thereto.
I can add text to the draft. Do you have ideas of what this text should say?
Also, it might be worth to have an socket API consideration appendix...
Yes, it would need a small API update to work - this can be described if
people want to progress this.
I think the main question we face is which approach to advocate:
* UDP
* An updated UDP with no checksum
* UDP-Lite
* A UDP-Lite profile (e.g. only minimal coverage)
* UDPTT
I don't know if I answered your questions.
Best wishes,
Gorry
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
